Spun out of MIT CSAIL, four founders - Ramin Hasani, Mathias Lechner, Alexander Amini, and Daniela Rus, started Liquid AI with the mission of building state-of-the-art, general-purpose AI systems that are capable, efficient, highly aligned, and trustworthy.
On this panel at Imagination In Action’s ‘Forging the Future of Business with AI’ Summit, Hasani and Amina discuss how the company thinks about AI and how they plan to make it more efficient.
Subscribe to FORBES: https://www.youtube.com/user/Forbes?sub_confirmation=1
Fuel your success with Forbes. Gain unlimited access to premium journalism, including breaking news, groundbreaking in-depth reported stories, daily digests and more. Plus, members get a front-row seat at members-only events with leading thinkers and doers, access to premium video that can help you get ahead, an ad-light experience, early access to select products including NFT drops and more:
https://account.forbes.com/membership/?utm_source=youtube&utm_medium=display&utm_campaign=growth_non-sub_paid_subscribe_ytdescript
Stay Connected
Forbes newsletters: https://newsletters.editorial.forbes.com
Forbes on Facebook: http://fb.com/forbes
Forbes Video on Twitter: http://www.twitter.com/forbes
Forbes Video on Instagram: http://instagram.com/forbes
More From Forbes: http://forbes.com
Forbes covers the intersection of entrepreneurship, wealth, technology, business and lifestyle with a focus on people and success.
On this panel at Imagination In Action’s ‘Forging the Future of Business with AI’ Summit, Hasani and Amina discuss how the company thinks about AI and how they plan to make it more efficient.
Subscribe to FORBES: https://www.youtube.com/user/Forbes?sub_confirmation=1
Fuel your success with Forbes. Gain unlimited access to premium journalism, including breaking news, groundbreaking in-depth reported stories, daily digests and more. Plus, members get a front-row seat at members-only events with leading thinkers and doers, access to premium video that can help you get ahead, an ad-light experience, early access to select products including NFT drops and more:
https://account.forbes.com/membership/?utm_source=youtube&utm_medium=display&utm_campaign=growth_non-sub_paid_subscribe_ytdescript
Stay Connected
Forbes newsletters: https://newsletters.editorial.forbes.com
Forbes on Facebook: http://fb.com/forbes
Forbes Video on Twitter: http://www.twitter.com/forbes
Forbes Video on Instagram: http://instagram.com/forbes
More From Forbes: http://forbes.com
Forbes covers the intersection of entrepreneurship, wealth, technology, business and lifestyle with a focus on people and success.
Category
🤖
TechTranscript
00:00Thank you, John.
00:04Hello, everyone.
00:05We are so happy to be here to tell you about our adventure in AI, which is called Liquid
00:13AI.
00:14So, let me get started by asking our CEO, Ramin Hassani, how did it all get started?
00:20Amazing.
00:21So, yeah, you guys hear me.
00:24Hi, everyone.
00:25Ramin Hassani, co-founder and CEO.
00:29You know, we invented a technology called liquid neural networks.
00:36This technology is actually inspired by biology, is a new type of AI system that is more explainable,
00:43that is more rooted in the math that we understand.
00:48So if you build basically large and large systems of this type of technology, the system
00:55is always fully understandable and we can really control.
00:59We have a lot of control on the design of these systems.
01:02They can solve problems in a very energy efficient way.
01:06How your brain is actually performing computation is actually, it doesn't consume the entire
01:11power of the United States to actually run, you know, or train or obtain these models.
01:17By the way, I've been running on a bar of chocolate today.
01:21That's true.
01:22Exactly.
01:23That's the amount of energy.
01:24And then, yeah, so everything started from Daniela's lab, where we invented the technology,
01:30like it's myself, Matthias Lechner, our CTO, who is not here today, with Daniela and Alexander,
01:36we co-invented this technology.
01:39And we tried over the last seven years, we tried to really bring it at a stage where
01:44it can solve real world problems.
01:47And we applied it in the space of autonomy, we applied it in the space of, you know, like
01:53modeling, like building forecasting models for markets, for medical data, and we soon
02:00realized that we have a technology that could be applied universally across kind of sectors.
02:06So we can apply to solve problems of sequential kind of nature.
02:10So if you have sequential data, like in the form of videos, audios, text, user data, financial
02:17time series, medical time series, and this kind of data, you can actually apply this
02:20technology and achieve state-of-the-art performance.
02:24Do that in a very, very computationally efficient way.
02:28And at the same time, running it on a math that you can completely understand.
02:33So these are the value props of the technology.
02:36So we decided to start the company with the idea of forming a new kind of foundation model
02:42for AI.
02:43Because these are elements that are missing right now from today's AI systems.
02:48So we want to have reliability, efficiency, and performance that is considerably good
02:53on tasks that matter for us like day-to-day.
02:56We want AI systems to solve problems for us in a way that we can control them.
03:01And not just being like a, you know, like today we have like productivity kind of tool
03:05systems, but we want to get them one step further to actually be able to solve problems.
03:11So as part of this panel, so we also invited some of our business development teams.
03:16So we have Nick Pagliuca with us, and we have Luis Hunt, both of them from business development
03:23team, and we have our chief strategist, AI strategist there, Joshua Bach, the legend.
03:32And as you know, all might know Alexander already, that's part of the founding team,
03:37and Daniela Rus.
03:39Do you want to tell them about the name, why it's called Liquid?
03:42Oh, yes.
03:43So maybe you can talk about it.
03:45Right.
03:47Okay.
03:48Liquid Neural Networks.
03:49So Liquid comes from this idea of being very flexible and adaptable.
03:52One of the key properties of our own biological intelligence is our ability to really be flexible
03:59in what we know, but not only what we know, but to go beyond what we know and extrapolate
04:04and adapt very easily to brand new domains that we've never been trained on before.
04:09And this is actually one of the hallmarks for what I think of as truly intelligent systems.
04:14It's the ability not just to be statistical models trained on a fixed data set, but actually
04:19take that data and try to learn some generalizable patterns.
04:22So Liquid really comes back to that idea.
04:25It gets at this concept of being very fluid, very adaptable in many of the properties that
04:32it has, and this gives it some nice explainability features as well that Ramin mentioned.
04:37But Liquid itself, it's also, it has this nice tie, not only the semantic meaning, that
04:43it's like a very flexible and adaptable system, but it also relates back to kind of the mathematical
04:48grounding of the Liquid model as well.
04:50There's a certain parameter in the Liquid model, which is called the Liquid Time Constant
04:54Parameter, and that's actually the real grounded origin of Liquid Neural Networks.
05:01Right.
05:02So let's not forget the hairy math that makes it all work.
05:05Okay.
05:06So I want to share a bit about how we think about solving the kind of problems Ramin referred
05:12to.
05:13Now, when we think about large foundation models that are being trained today, they're
05:22extraordinarily energy hungry.
05:24They cost a lot in terms of energy, they need huge server farms, and they cannot run on
05:33edge devices nor on company enterprise computers.
05:38And so how does Liquid AI think about this problem and how does Liquid AI work towards
05:47more practical solutions as far as energy is concerned?
05:52Yeah.
05:53So what we're building in-house is, can you hear me?
05:59Yeah.
06:00So what we're building in-house is a developer package where we give enterprises access to
06:08this technology to solve problems.
06:11Now these problems could be domain-specific AI problems and generative AI problems.
06:17So in both kind of domains, we have been very successfully interacting with many clients
06:21already, and maybe Luis, you can talk a little bit about the business angle of things.
06:28But in terms of how we are bringing value is giving clients basically the opportunity
06:34to work with a private version of AI models and best-in-class kind of AI models.
06:42So we give the technology all together in this developer package to companies to build
06:47use cases for.
06:48Now the properties that I told you that you would get as a result of using Liquid AI's
06:53technology is the fact that you're not going to need a massive amount of compute to actually
06:58build AI solutions.
07:00For example, for generative AI applications, you can build Liquid foundation models as
07:07opposed to GPTs that are between 10 to 20 times more cost-efficient and energy-efficient
07:15in-house if you use the foundation of Liquid.
07:19Now that comes with the explainability and all the features that I said.
07:24You can get the state-of-the-art performance.
07:27At the same time, when you actually take the technology and deploy it for inference
07:31and usage of the technology, you can gain between 10 to 1,000 times faster inference
07:38time.
07:39That's going to be much cheaper to execute if you have generative AI solutions.
07:43So in some sense, you're providing access to a lower carbon footprint.
07:48So development of very powerful AI systems today with the Liquid AI technology is possible
07:54with a fraction of the cost that the companies, the big tech is actually today spending.
08:00And that was one of the core properties of our technology.
08:05Maybe you can talk a little bit about our business approach, how we are going and how
08:08we are looking.
08:09Yeah, actually.
08:10So given this extraordinary benefit that would put the technology in the pockets, basically
08:18on the enterprise computers of companies, how do you think about helping companies with
08:22the AI transformation?
08:24Sure.
08:25Well, just the notion in general of this idea of a smaller, more performant, more powerful
08:29neural network model that companies can bring in-house so that the data never has to leave
08:34their premise, never has to leave their environment.
08:37The models never have to leave their environment.
08:39I know Danielle, as the director of MIT CSAIL, has been getting inbound requests for something
08:43like that for a long time.
08:44But it doesn't just stop there in terms of the genius of the brilliant AI scientists
08:49on the team is that they've not only created a developer package that allows all those
08:55years of research to create these incredibly powerful sort of groundbreaking models, to
09:00allow the ML and AI scientists at enterprises all over the world to be able to build those
09:06kinds of models using the package in a way that's seamless, in a way that's easy to implement.
09:11And so now enterprises have the ability to be able to customize those models to their
09:15specific needs, whether it's in finance and whether it's an algorithmic trading firm that
09:19needs to be able to have sort of stock price predictions every half second with very high
09:24levels of accuracy, or whether it's another financial service firm that needs to be able
09:28to do anomaly detection or credit card fraud detection at a rate of 3,000 transactions
09:32per second or faster, or whether it's firms in bio, pharma, and life sciences that need
09:39a superior ability to do molecular sequence modeling.
09:43And so the use cases abound.
09:46It's a matter of focus now more than in prioritization, if anything.
09:50But we're excited to start to get this remarkable developer package and these models in the
09:55hands of more companies around the world.
09:58And Nick, I wonder if you can add to that and share some example business applications
10:03we're actually working on.
10:05Yeah.
10:06So Louis mentioned a few of them.
10:07So within financial services, the types of time series data that is very common work
10:12really well with these models.
10:14And so a couple of the examples Louis mentioned, either stock price prediction or fraud detection
10:19are really key value drivers.
10:22We're really trying to get away from some of the hype in the AI space and focus on problems
10:28that actually impact businesses.
10:31And so the best way to do that is to work directly with these enterprises and understand
10:35their problems and then map the value of these models, which it may be efficiency, it may
10:40be performance, it may be latency to those types of use cases.
10:46Yeah.
10:47So that's so exciting.
10:49And I guess what I want to underscore is that it's not just about the fact that the model
10:54is smaller.
10:56The model is also more accurate.
10:58And so I wonder if one of you, maybe Ramin, you can tell us how are you thinking about
11:07the main challenges of improving accuracy?
11:10What exactly can we do in order to enhance the accuracy of the models?
11:17Because we don't want to have models that hallucinate.
11:21And especially for safety critical applications, we cannot have AI that makes mistakes.
11:27If there's an AI on a self-driving car and it doesn't detect pedestrians, that is a big
11:32problem.
11:33How are we avoiding that?
11:34Yeah.
11:35Great question.
11:36So whenever we want to bring a technology in real world, we need to ensure that the
11:40technology is reliable.
11:42It's accurate and it's reliable.
11:44Now today's generative AI solutions are not reliable.
11:47The reason being, and as a matter of fact, they're not getting adopted because of this
11:52reliability issue.
11:54And that Daniel refers to as the accuracy and performance of the models.
11:59So think about how we are designing generative AI today.
12:04We train large neural networks on massive amounts of data to predict the next token.
12:12So that means you're designing a hallucination engine, basically.
12:16That's the process.
12:17You're building a system to hallucinate.
12:19Why?
12:20Because you want to create a system that has incorporated the knowledge of the world in
12:25that sense.
12:26So now, this is the first phase of obtaining a model.
12:30The model is a hallucination engine, basically, because it predicts the next token all the
12:35time.
12:36Then you go through a fine-tuning kind of phase.
12:39Fine-tuning phase is getting this model to really fit the data that you want to see.
12:46These are labeled data.
12:48Fine-tuning process is basically labeled data when you want to have an input, question answering,
12:53for example, is one of those schemes.
12:55You want to guide this hallucination engine to answer some questions for you.
13:01This is a process where you're trying to reduce the hallucination rate of the underlying technology,
13:05which is a transformer architecture.
13:08Now, transformer architecture itself is a feed-forward architecture.
13:12So if you have a feed-forward system, that means it receives an input and it generates
13:16an output.
13:17So there's no feedback mechanisms involved.
13:20One of the things that our technology is actually bringing forward is the fact that we can now
13:25design feedback systems at scale.
13:29So that means there's only that much that with a feed-forward system you can actually
13:33guide the hallucination of a network.
13:36If you have a feedback system, like a human, your brain is actually wired in a very condensed
13:42and recurrent way.
13:44That allows you to actually control the hallucination rate much better.
13:49So this is actually how we are fundamentally trying to resolve the hallucination problem
13:54of these models.
13:56Ramin loves to talk about the brain and the company aspires to have an important role
14:03in AGI, in Artificial General Intelligence.
14:07And so I want to come to Joshua and ask you, what do you see as the scientific and technological
14:14challenges that will take us from where we are now to AGI?
14:20And also, do you see any specific milestones that we have to hit that will actually tell
14:26us that we're on a good path?
14:28Well, if I could tell you how to get to AGI, Ramin would have all your sign in NDA first.
14:37So this is really an open question.
14:38It's also what makes this so exciting.
14:40What is the delta between the performance of the system that we are seeing and the capabilities
14:45of a human being in all relevant dimensions?
14:50What we find is, for instance, our mind can hallucinate, but we can also make proofs.
14:55We can ground our perception in what we are seeing and what we are perceiving and in our
15:00motivational system.
15:01So we have a certain contextualization that is both given by our empirical access to the
15:06world and by our epistemology, by our ability to decide what's true, what's factual, and so on.
15:12And it's at this point not clear if the present architectures can solve this by scaling them
15:17up and getting better by introducing more and more data into the model and using more
15:21compute, or whether we have to build systems that are working more like our mind is doing,
15:27which means we are entangled with the world.
15:29We are always getting feedback from the world in which we are embedded in.
15:32We continuously adapt our architecture, and not just to the world, but also to ourselves.
15:37We basically are self-improving.
15:39And so I think one of the challenges is to reflect on and observe how biological systems
15:46are operating when they are solving problems and incorporate many of these ideas.
15:51But this does not mean that the present approaches are doomed.
15:53We just don't know.
15:54There are a lot of people which bet on the scaling hypothesis that you can basically
15:58brute force your way to a system that is so smart that it's able to help us building better,
16:03more elegant systems.
16:04We all know that the present architectures are too energy inefficient.
16:08We know that they need too much data to train.
16:10But we also see that they are unlocking capabilities that are very often surprising and amazing.
16:15For instance, there is no system that is able to demonstrate learning language like a human
16:20child by pointing at things interacting with the world.
16:24But once we train an LLM up, we can give it a language it has never seen and put it into
16:29its prompt context, and it learns from much less data a new language than a human being
16:33could.
16:34So in some dimensions, these models are already superhuman.
16:37But in other levels, they are not human and intelligent at all.
16:41And so it's this very interesting gap now where we are able to develop tools that are
16:45super powerful that we just begin to learn how to use and integrate into our world, into
16:49our lives, but that are not AGI.
16:52And the only AGIs are also not we as individuals, but it's a civilization that needs to spend
16:58many, many generations before we can bootstrap ourselves collectively to the level that we
17:03begin to make sense of reality.
17:05And as you think about the special properties of the liquid models, what are you betting
17:13on?
17:14What are you excited about?
17:15What I find especially fascinating about liquid models is they do not rely on a fixed architecture
17:21where you start out with a predefined thing and use the perceptual algorithm is very powerful.
17:27But the main reason why everybody is using it for so long since the 60s is because it's
17:31also very simple to understand and very simple to build something that can train on it.
17:37And when we look at the way in which a brain is operating, it's much more fluid.
17:41And so if we go back to the drawing board, as Ramin did in his PhD and Daniela Skup did,
17:48we are discovering ways of making this more elegant.
17:52And while the mathematics looks on the surface slightly more hairy, you soon realize that,
17:56oh, it's actually a lot less work to do to get your system to do the same stuff as before.
18:02And so there is, I think, an important step towards systems that are more fluid and are
18:05more adaptable to the tasks and to the environment that they're in.
18:09I mean, in traditional AI systems, the neuron is essentially an on-off system, right?
18:15It's zero or one.
18:16And you have lots and lots of them and some very basic math that makes information flow
18:22between them.
18:24But that's not how nature works.
18:26Yes.
18:27And so a neuron is not just a sum over a few real numbers that come in and a non-linearity.
18:32And if then thrown in, basically, it's a little organism that is a reinforcement learner that
18:37can solve complicated problems in a continuous way.
18:41And by incorporating some of these ideas, we get a lot more powerful architectures.
18:46And it's basically what makes this very exciting from a research perspective to be part of
18:50this.
18:51You know, John is not here, so I think we can keep going.
18:56So, what is next for Liquid?
19:03Yeah.
19:04So, I mean, one of the things that I want to do, like, I mean, the technology itself
19:09is like one of those, you know, out of the box thinking kind of ways.
19:13And it also induces hype, you know, we are also aware of this thing.
19:16But one of the things that at the company, we've been trying to keep our heads down.
19:20And you know, like, if you look at our accounts, like what we are announcing in the social
19:23media, it's basically like very minimal kind of information we are releasing.
19:28We're just absolutely going to bring value and you're making this thing real.
19:32And I think this is one of the things that at the company, the last year, we've been
19:37very silent, you know, and we have been like operating on a, you know, like delivering
19:42value to our clients.
19:43And this has been very, very rewarding journey for us.
19:47Like, we have been, you know, one of the few AI companies that have already have like interacted
19:52with like, you know, 50 plus kind of Fortune 500 companies, which are already using the
19:57technology and are in negotiations with the company to actually integrate this technology
20:02in their work.
20:03And we are just keeping our heads down and executing and bringing the value of AI to
20:08enterprises.
20:09Well, since John is not here, I'm sure that everyone will want to be very clear on what
20:16the company offers.
20:17So let's say I'm a new customer looking to expand my AI know-how and looking to bring
20:26foundational models in my business.
20:29What can Liquid AI give to me?
20:32Yeah.
20:33Maybe so.
20:34Two levels, right?
20:35Level number one is it gives you a comprehensive developer package and infrastructure around
20:40that to deploy, to develop and deploy in-house liquid neural networks that are capable and
20:47private.
20:48So capable in terms of being highly accurate, but private being in terms of, you know, small
20:52enough that you could have your own internal massively performant AI model entirely in-house.
20:58And then on the other side, you know, Liquid AI offers the actual models themselves, not
21:02only a developer package, but actually the models and the infrastructure around those
21:05models to, number one, you know, train, fine-tune and deploy on your own data as well.
21:12So you kind of have these two sides of the coin with what we offer to our customers.
21:18On one side, it's a very low-level developer package.
21:20On the other side, you know, the models that can actually augment all of those functionalities
21:26for customers.
21:27Well, thank you for the opportunity to tell you about our exciting adventure.
21:34And thank you to our wonderful panellists and leaders in AI.
21:39Hold on, hold on, hold on, hold on, hold on, hold on.
21:44In just closing, is there an ecosystem that St. Louis was Silicon Valley before Silicon
21:51Valley?
21:52Detroit was Silicon Valley before Silicon Valley.
21:55Actually, Buffalo was offered the car industry and the plane industry, and they said no,
22:01and it went to Detroit.
22:03Buffalo could have been Silicon Valley.
22:05And in the 1960s, there was a list of two cities that weren't going to make it, Detroit,
22:09and we know what happened there, you know, in terms of economics.
22:11I really appreciate, Randall, you have been propping up Detroit, doing some amazing stuff.
22:17Boston and Detroit were two cities that weren't going to make it.
22:21Boston's done okay.
22:23We've reinvented ourselves.
22:24Could what Liquid has created create a seismic change of the geography of where innovation
22:34and economies and opportunities, like, just say Silicon Valley, watch out, and the rest
22:41of the world, watch out.
22:42Like, is there an opportunity?
22:43Just go.
22:44One comment on this.
22:45Go, each of you.
22:46I still live in Silicon Valley, because there are a lot of ideas that exist there.
22:51And while I love Boston, I felt that AGI was always a little bit further in the future
22:56in Boston than it was in Silicon Valley.
23:00Not as much as in Berlin or Zurich, where it's basically never going to happen.
23:03But I think that this is an extremely important place.
23:09Like, MIT is not necessarily always the place where you build products.
23:13It's the place where you build the ideas that will drive the products.
23:17This is the amazing value of this place, that you have people who actually think, who do
23:22not just try to apply what others have thought about.
23:24You're a little theoretical.
23:25I'm asking, can Liquid change things?
23:27All right.
23:28Liquid.
23:29Business development guy.
23:30Can Liquid make a difference?
23:31I think that whenever you have a foundation of talent like you have at MIT, and then you
23:35have an incredible community at MIT, as well as folks that help to foster and cultivate
23:39that community, you, John, Dave, just extraordinary, both in terms of building the community, also
23:44supporting companies like us.
23:47The transistor, going from tubes to transistors, that created a new ecosystem.
23:51I'm saying you guys are on to something big.
23:54All right.
23:55What do you think?
23:56Kid who just graduated with honors from Duke in computer science, didn't do Latin or Greek,
24:01and then with distinction from HBS.
24:04What do you think?
24:05You're like, you don't have gray hair yet.
24:06No, I do not.
24:07A little bit.
24:08But, I mean, innovation can happen anywhere.
24:11MIT is a breeding ground for innovation.
24:14And Boston is as well.
24:16So, I'd say innovation can happen anywhere.
24:18And we are just trying to put ourselves in the position to be people innovating.
24:21And we have a great team of researchers led by these two, and this one, and a bunch of
24:26people out of MIT, but a bunch of people out of other universities that are also really
24:30...
24:31All right.
24:32You three have done good, but co-founders, just, or tri-founders, can you say, how can
24:36Liquid create an ecosystem that will change the vector of society?
24:41What do you got?
24:42So, very simple from my end.
24:43I think that the company is not done innovating.
24:45I think that we still need a ton of innovation to get to AGI, similar to what Yoshua said.
24:49I think that not only we need innovation, I think we need out-of-the-box thinking.
24:52It's not just about implementing existing ideas.
24:55We need to reinvent brand new things and go back to first principles in order to do that.
24:59And there's no better place, in my opinion, than the MIT community to nurture that.
25:05All right.
25:06Sio, what do you got?
25:07Yeah.
25:08I mean, we could always praise ourselves all the time, but we can also learn.
25:10So, one of the things that we have actually achieved at Liquid is, one of the few companies
25:15that you're having massive amount of superstars from Stanford joining a Boston headquartered
25:20kind of company.
25:21So, this is one of the ...
25:22It's like sea turtles.
25:23Very, very, very ...
25:24Throwing out, coming back.
25:25Exactly.
25:26It's like the other way around.
25:28And I think this just comes with the fact that it is not all about ... Of course, you
25:33have a great idea, but you need to have a plan.
25:36How do you want to execute that plan and stuff?
25:38And I think everything, for us, is like, we have a foundation model that is ... I mean,
25:43as I said, I don't want to contribute to the hype around everything, because everything
25:48I say, it might actually be interpreted like that.
25:50But as I said, we're keeping our heads down.
25:55We're attracting the most important talent in the world that are actually innovating.
25:59You can see, we have 30 people at Liquid right now that they're the brightest minds on earth.
26:04This is such an energizing kind of experience for myself, as the CEO of the company.
26:14Surrounding myself with amazing people that can contribute to this core, that we want
26:18to build an AGI that we can understand.
26:21We want to build AGI, and it's going to be this company that is actually going to ... This
26:27is what I believe, that we are onto something crazy here.
26:31But it is going to take time.
26:33It is going to take talent, and apart from that, there is also the support of ... Silicon
26:38Valley became Silicon Valley because there are massive amount of capital and business
26:42kind of surrounding the Stanford community actually went into the whole thing.
26:46So I think that's the part that we have to get a little bit moving.
26:51We have to be a little bit more risk taking, as people of Massachusetts, I would say.
26:56And that's the part that I think we've been successfully igniting something here, and
27:01we are on the virtue of making it even more.
27:03But yeah, bringing both business and the technology thing together to make it happen.
27:09All right, in closing, she's a MacArthur Genius Fellow, but forever she will be known as the
27:14TED homepage speaker that launched on Friday.
27:18The answer I was looking for, which none of them got, was we're going to be oxygen for
27:22a lot of ventures.
27:25People are going to say, look what happened because we created something on the other
27:28side of firewalls, on robots, on using less energy.
27:34And these guys just built something that I'm so excited about.
27:37All right.
27:38Moderator, Ms. Moderator, what do you got?
27:40Well, let me just say that ideas, and especially out of the box ideas, change everything.
27:46And Massachusetts and the Boston area has 100 universities with so many bright minds
27:53who can change everything.
27:54And so I live in Boston, and I believe that this is a great place for AI and for the next
27:59AI revolution.
28:00All right.
28:01Get off, get off, get off, get off.
28:24Get off, get off, get off, get off, get off, get off, get off, get off, get off, get off,
28:33get off, get off, get off, get off, get off, get off, get off, get off, get off, get off,
28:40get off, get off, get off, get off, get off, get off, get off, get off, get off, get off,
28:46get off, get off, get off, get off, get off, get off, get off, get off, get off, get off,
28:52get off, get off, get off, get off.